tgalante

New Member
Hi All,

I am trying to use a complex ANOVA design with the Permanova function in PRIMER. I have been working on this for a while with no significant progress, so any ideas or help would be greatly appreciated.

The SAS code for my design is:

proc mixed data=treatment;
class Gap Wood Collar Year Month Plot;
model Biomass= Gap|Wood|Collar|Year|Month / outp=Biomass2;
random plot(Gap*Wood) plot(Gap*Wood*Collar) plot(Gap*Wood*Collar*Year);
run;

Resulting in denominator DF of 16 for Gap, Wood, Gap*Wood; 32 for collar and collar* combinations of gap and wood; 48 for year and combinations of gap and wood and collar; and 93 for everything else.

The problem I am having in PRIMER is that I can only nest plot in one combination of these terms, resulting in denominator DF that are too large. I think that the answer to this question has something to do with pooling terms. Any help or direction that you can give would be greatly appreciated. Thanks.

Sincerely,
Tera

bugman

Super Moderator
Tera

before I can help you, would you please detail your design and your factors (sample size and whether you are treating them as random or fixed) and tell me what you are trying to do?

tgalante

New Member
Hi Bugman,

Sure, I can give you more information. There are 6 factors in my design: Year, Month, Gap, Wood, Collar, and Plot. The treatment factors are a 2 x 2 factorial design of gap and wood; canopy gap added or not to a plot, wood added or not to a plot. This results in 4 treatment levels. Each treatment is replicated in 5 plots, resulting in 20 plots total. Within each plot, soil was sampled at 3 locations, which I call collars. This sampling was done in two years, and two months within each year. All of the factors are fixed except plot.

Plots have 4 levels of nesting, within: (gap x wood), (gap x wood x collar), gap x wood x collar x year), and (gap x wood x collar x year x month) (I think that this last level of nesting is not truly nested, rather it is using the residual df as the denominator. This ANOVA design results in distinct denominator df for effects containing gap or wood (den. df=16), anything with collar (den. df=32), anything with year (den. df=48), and anything with month, or everything else (den. df=93). Therefore, this design allows me to test each level of nesting with the correct denominator df.

If this is not possible in PERMANOVA, I would accept a test that was more conservative. For instance, when I nest plot within all of the other factors in PERMANOVA, I obtain a very large den. df for each factor. The manual states that a larger den. df is giving me higher power, and more probability of rejecting the null hypothesis when it is false. I still think that the answer to this lies in pooling factors, but have not figured out exactly how to do this.

Thanks for reading. I really appreciate any help on this.

Sincerely,
Tera

bugman

Super Moderator
Tera

I apologise for letting this thread go cold. I got swamped at work. Let me know if you still need help (but by the way i dont think its a pooling issue) - I think its realted to how you have treated each factor in the initial model.